Wednesday, May 27, 2009

Chrysler Dealers: No Correlation at State Level

There is no correlation of Chrysler Dealer closings at the state level.

I began my analysis by first downloading the "cut" list and "kept" list in PDF format. I reformatted those into a spreadsheet (in MS Excel). Next, I got a copy of the state-by-state returns for the 2008 election from wikipedia and added those to the spreadsheet. This allowed me to compare the ratio of Obama vs McCain to the ratio of kept vs cut dealers. You'll find those numbers on the "analysis" tab of the spreadsheet. My thinking was that there would be a direct correllation between how strongly a state supported Obama and how many dealers it lost.

There were no dealers in DC, so I threw it out. I believe that states with ten or fewer total dealers (cut plus kept) are not relevant statistically, so I also took out DE, HI, RI, and AK. With that I got a coefficient of corelation of 0.1361. A value close to 1.0 would indicate a strong correlation—Obama favored states that supported him. A value close to zero indicates that there is no correlation. And a value close to -1.0 would imply that Obama favored states that opposed him.

For the ten states with the largest total number of dealers (TX, PA, IL, NY, MI, OH, CA, FL, WI, and IN) the coefficient is higher: 0.3069. However, if we extend that to twenty states (adding MO, IA, MN, NC, NJ, VA, GA, OR, TN, and KY), the coefficient goes negative: -0.1049. It's really a mixed bag.

I'm not a statistician, so there are may be problems with my approach. Leave your corrections and suggestions in the comments! And, please download the spreadsheet (MS Excel) and perform your own analysis! I've put the zipcodes in their own column; however, I do not have county/zipcode level voting data. I'm hoping someone out there might be able to either email that to me (dsm012 -at- gmail -d0t- com) or do the analysis themselves and leave a comment with a link to their work!

Now to look at the OpenSecrets donor data...

Make sure you checkout the Doug Ross's latest work.

Update: After my initial post I remembered that I had omitted the first page of data from the "kept" PDF. It was formatted differently so my plan was to incorporate it last. This data does not change my belief that there is no correlation at the state level. I have corrected the numbers above to reflect the new data.

Update 2: Check-out my analysis of urban vs rural closures!


Sue Sarkis said...

I took the time to check the political contributions of 60 of the dealerships that were forced to close and ALL 60 were very heavy GOP contributors. Someday maybe I'll have time to check more.

dsm said...

Sue, keep digging! I think there's more to this story, but, well, it's been eclipsed by other issues (cap-n-tax and healthcare). That said, you need to be careful about the approach you use to analyze the data. It's fairly easy to find sixty, even a hundred, dealers that were big republican donors. Did you randomly choose them? If so, what mechanism did you use.

I think the administration was shrewd enough to make the aggregate numbers look fair. However, I think they settled a few political scores with specific political opponents. In other words, don't look at all of the closings. Rather, try to dig into the specifics of a few of them.